Professional Documents
Culture Documents
Lecture 16
Controllability and state transfer
EE263 Autumn 2003
• state transfer
• reachable set, controllability matrix
• minimum norm inputs
• infinite-horizon minimum norm transfer
• controllability Gramian
Controllability and state transfer 16 – 2
State transfer
questions:
• where can x(ti) be transfered to at t = tf ?
• how quickly can x(ti) be transfered to some xtarget?
• how do we find a u that transfers x(ti) to x(tf )?
• how do we find a ‘small’ or ‘efficient’ u that transfers
x(ti) to x(tf )?
Controllability and state transfer 16 – 3
Reachability
• Rt is a subspace of Rn
• Rt ⊆ Rs if t ≤ s
(i.e., can reach more points given more time)
t≥0
Controllability and state transfer 16 – 4
· ¸
t−1
where Ct = B AB · · · A B
thus we have
range(Ct) t < n
Rt =
range(C) t ≥ n
where C = Cn is called the controllability matrix
0 1 1
example: x(t + 1) = x(t) + u(t)
1 0 1
1 1
controllability matrix is C =
1 1
i=0
then
n−1
Az = Ai+1Bu(i)
X
i=0
n−1
= AiBu(i − 1) + AnBu(n − 1)
X
i=1
n−1
= AiB(u(i − 1) − αiu(n − 1)) − α0u(n − 1)B
X
i=1
∈ range(C)
where
X (s) = sn + αn−1sn−1 + · · · + α0 = det(sI − A)
and we have used the C-H theorem:
An = −α0I − · · · − αn−1An−1
Controllability and state transfer 16 – 7
x(tf ) = A tf −ti
x(ti) + Ctf −ti
.
.
u(ti)
τ =0
is given by
uln(t − 1)
..
T T −1
= Ct (Ct Ct ) xdes
uln(0)
for τ = 0, . . . , t − 1
Controllability and state transfer 16 – 9
t−1
Emin, the minimum value of ku(τ )k2 required to reach
X
τ =0
x(t) = xdes, is sometimes called minimum energy
required to reach x(t) = xdes
t−1
Emin = kuln(τ )k2
X
τ =0
µ ¶T
= T T −1
Ct (CtCt ) xdes CtT (CtCtT )−1xdes
= xTdes(CtCtT )−1xdes
t−1
−1
T τ T T τ
= xdes A BB (A ) xdes
X
τ =0
Emin as function of t:
if t ≥ s then
t−1 τ T T τ s−1
A BB (A ) ≥ Aτ BB T (AT )τ
X X
τ =0 τ =0
hence
t−1
−1
s−1
−1
τ T T τ τ T T τ
A BB (A ) ≤ A BB (A )
X X
τ =0 τ =0
so Emin(xdes, t) ≤ Emin(xdes, s)
1.75 0.8 1
example: x(t + 1) = x(t) + u(t)
−0.95 0 0
Emin(z, t) for z = [1 1]T :
10
6
Emin
2
PSfrag replacements
1
0
0 5 10 15 20 25 30 35
t
0
x2
−5
−10
−10 −8 −6 −4 −2
x10 2 4 6 8 10
10
Emin (x, 10) ≤ 1
5
PSfrag replacements
0
x2
−5
−10
−10 −8 −6 −4 −2
x10 2 4 6 8 10
Controllability and state transfer 16 – 12
the matrix
t−1
−1
τ T T τ
P = lim A BB (A )
X
t→∞ τ =0
¯
t−1 ¯
ku(τ )k2 x(0) = 0, x(t) = xdes = xTdesP xdes
¯
min
X ¯
¯
¯
τ =0 ¯
t−1
then Aτ BB T (AT )τ converges as t → ∞ since Aτ
X
τ =0
decays geometrically
the matrix ∞
Wr = Aτ BB T (AT )τ
X
τ =0
is called the reachability (or controllability ) Gramian
Continuous-time reachability
therefore
Z
t Aτ
x(t) = 0 e Bu(t − τ ) dτ
Z
t n−1 i
= α Bu(t − τ ) dτ
X
0
i (τ )A
i=0
n−1 i
Z
t
= A B 0 αi(τ )u(t − τ ) dτ
X
i=0
= Cz
Z
t
where zi = 0 αi (τ )u(t − τ ) dτ
Impulsive inputs
then U (s) = sk f , so
X(s) = (sI − A)−1Bsk f
µ ¶
= s I + s A + · · · Bsk f
−1 −2
hence
k k+1 t k+2 t2
x(t) = impulsive terms +A Bf +A Bf +A Bf +· · ·
1! 2!
in particular, x(0+) = Ak Bf
by linearity we have
f0
example
• two unit masses with positions y1, y2, connected by
unit springs, dampers
• input is tension between masses
• state is x = [y T ẏ T ]T
u(t) u(t)
PSfrag replacements
system is
0 0 1 0
0
0 0 0 1 0
ẋ = x + u
−1 1 −1 1 1
1 −1 1 −1 −1
controllability matrix is
0 1 −2 2
0 −1 2 −2
· ¸
2 3
C = B AB A B A B =
1 −2 2 0
−1 2 −2 0
where
hA
Z
h τA
Ad = e , Bd = 0 e dτ B
Controllability and state transfer 16 – 21
similarly
NX
−1 NX
−1 Ati/N T ti/N
AidBdBdT (ATd )i = e BdBdT eA
i=0 i=0
t At̄ T AT t̄
Z
≈ (t/N ) 0 e BB e dt̄
for large N
min energy is
Z
t 2
0 ku ln (τ )k dτ = xTdesWr (t)−1xdes
where
t At̄ T AT t̄
Z
Wr (t) = 0 e BB e dt̄
can show
(A, B) controllable ⇔ Wr (t) > 0 for all t > 0
⇔ Wr (s) > 0 for some s > 0
in fact, range(Wr (t)) = R for any t > 0
Controllability and state transfer 16 – 23
( Z ¯ )
t 2
= xTdesP xdes
¯
min 0 ku(τ )k dτ ¯
¯
¯ x(0) = 0, x(t) = xdes
Reachability Gramian
since
A(tf −ti )
Z
tf
x(tf ) = e x(ti) + ti eA(tf −τ )Bu(τ ) dτ
example
replacements u1 u1 u2 u2
ẏ
system is:
0 0 0 1 0 0
0 0
0 0 0 0 1 0
0 0
0 0 0 0 0 1
0 0 u1
ẋ =
x +
−2 1 0 −2 1 0 1 0 u
2
1 −2 1 1 −2 1
−1
1
0 1 −2 0 1 −2
0 −1
Z
tf
Emin = 0 kuln(τ )k2 dτ vs. tf :
50
45
40
35
30
Emin
25
20
15
PSfrag replacements 10
0
0 2 4 6 8 10 12
tf
for tf = 3, u = uln is:
5
u1 (t)
−5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
5
t
PSfrag replacements
u2 (t)
−5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
t
and for tf = 4:
5
u1 (t)
−5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
5
t
PSfrag replacements
u2 (t)
−5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
t
Controllability and state transfer 16 – 28
output y1 for u = 0:
1
0.8
0.6
y1 (t)
0.4
0.2
0
PSfrag replacements
−0.2
−0.4
0 2 4 6 8 10 12 14
t
output y1 for u = uln with tf = 3:
1
0.8
0.6
y1 (t)
0.4
0.2
0
PSfrag replacements
−0.2
−0.4
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
t
output y1 for u = uln with tf = 4:
1
0.8
0.6
y1 (t)
0.4
0.2
0
PSfrag replacements
−0.2
−0.4
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5